University of Toronto · Huawei Noah's Ark Lab Researcher

Hi, I'm Zhiyuan Li

Undergrad student with passion in mathematics

BASc Engineering Science (Robotics) · GPA 3.90 · UofT 2022–2027

🎓 Seeking Master / PhD (Fall 2027)

About Me

I'm a Robotics Engineering student at the University of Toronto, passionate about embodied AI, mathematics, and creative arts. Currently conducting research at Huawei Noah's Ark Lab on Vision-Language-Action models while building full-stack applications at Prelude CA. Award-winning mathematician (Putnam Top 154, UofT Math Contest #1) and aspiring music composer.

Zhiyuan Li

Research Experience

Research Assistant
Huawei Technologies, Noah's Ark Lab - Embodied AI
May 2025 - Present
Toronto, Ontario
  • Conducted research on Vision-Language-Action (VLA) models such as Pi0 for robotic manipulation in visually cluttered environments, studying perception-action coupling and robustness degradation. Designed experiments using UR5e and Franka robots with ROS integration.
  • Explored long-horizon symbolic planning via Plan-to-Evolve (P2E) frameworks. Fine-tuned LLMs including Qwen3 and LLaMA 3/4 using PDDL-based domains with Chain-of-Thought reasoning. Integrated real-robot experiments combining VLA and motion planners.
  • Contributed to Mem2Ego, a global-to-egocentric memory architecture for long-horizon embodied navigation. Integrated V-SLAM and Nav2 within ROS2 for agent-guided navigation.
  • Lead engineer on The North Star team at NeurIPS 2025 Behavior Challenge. Designed FSM pipeline integrating SAM6D, Pi0 VLA, and classical robot motion planning for 50+ household tasks in Isaac Sim. Achieved 4th place.
Undergraduate Research Assistant
UofT Continuum Robotics Laboratory
Feb 2024 - Apr 2025
Toronto, Ontario
  • Conducted in-depth research on state estimations for continuum robotics, focusing on constraints and degrees of freedom. Extended Clarke transformation to general models for tendon-driven continuum robots (TDCR).
  • Analyzed TDCR forward kinematics from first principles, evaluating hardware factors such as material stiffness, size effects, disk radius, and tendon length. Introduced corrective factors to improve model accuracy.
  • Developed and simulated TDCR models using MuJoCo and Pyplot. Documented findings through blog posts on the OpenCR project.

Publications

Self-CriTeach: LLM Self-Teaching and Self-Critiquing for Improving Robotic Planning via Automated Domain Generation
H-WM: Robotic Task and Motion Planning Guided by Hierarchical World Model
ICLR 2026 Workshop on World ModelsarXiv:2602.11291
Distracted Robot: How Visual Clutter Undermines Robotic Manipulation
Improving Robotic Manipulation Robustness via NICE Scene Surgery
Data and Evaluation Protocols for Robust Robotic Manipulation
RSS 2026 WorkshopProject
Do World Action Models Generalize Better than VLAs? A Robustness Study
Mem2Ego: Empowering VLMs with Global-to-Ego Memory for Long-Horizon Embodied Navigation

Awards & Honors

Putnam Math Contest

Score 42, Top 154 Individual, Top 3 member of Top 5 Winning Team

Canadian Mathematics Olympiad

Invited participant

UofT Undergrad Mathematics Contest

#1 among all participants from all years, all campuses

CEMC Math Contests

Euclid Honor Roll (Rank 77), Hypatia, Fermat, CSMC Honor Roll (Rank 100)

AMC/AIME Math Contests

2-time AMC Honor Roll (Top 0.1%), 2-time AIME Honor Roll (Top 0.1%)

Industry Experience

Full-Stack Engineer / Frontend Lead
Prelude CA
June 2025 - Present
Toronto, Ontario
  • Led the frontend architecture of a multi-tenant SaaS platform using Next.js 15 (App Router), React 19, and TypeScript. Built a modular design system with Radix UI and Tailwind CSS, implemented global state management via Zustand, and optimized performance through client-side state caching and cookies, reducing interaction latency by 40%.
  • Implemented secure authentication and authorization for workspace-isolated tenants, using OAuth integration (Google/Microsoft), role-based access control, and team invitation flows. Designed multi-tenant isolation across application and data layers.
  • Designed an AI-driven pipeline enabling automated, multi-agent data processing workflows. Built an intelligent CSV ingestion system with semantic column matching using OpenAI embeddings, achieving 90%+ auto-mapping accuracy.
  • Designed a scalable CRM relational schema using PostgreSQL, unifying customers, leads, and deals into comprehensive user profiles with normalization, indexed key columns, and transaction-safe updates.
Full-Stack Engineer
Yuhang International Flight (YIF)
July 2022 - May 2024
Dalian, China
  • Designed and developed an online information management system using Python, QtDesigner, SQL, and QML to track and manage IOU payments, significantly reducing manual input and improving efficiency.
  • Utilized MySQL with connection pooling for multiple concurrent users. Enhanced user authentication with bcrypt hashed passwords, enforcing security protocols.
  • Implemented advanced multi-key search using SQL JOIN, INDEX, and LIKE statements. Developed fuzzy search with FuzzyWuzzy library for partial matches. Integrated real-time updates with SQL triggers.
  • Leveraged pandas and numpy to connect to Excel and other platforms, enabling data import and visual summaries through graphs depicting IOU amount changes over time.
  • Conducted rigorous system testing with over 1 million items and 100+ concurrent users. Optimized to process 100k+ IOUs with 3-5 second search response, improving efficiency by over 100%.

Personal Projects

Desktop Mechatronic Robotic Pet

Multi-modal interactive robot using ESP32-WROOM microcontroller programmed in Arduino IDE and C++. Features multi-threaded motor control, Python API for desktop integration, and custom CAD-designed mechanical components. Supports voice commands, visual recognition, and wireless control.

Delivery Turtle-bot

ROS-based delivery robot programmed in Python with PID control for precise movement and Bayesian localization for accurate positioning. Implements autonomous navigation algorithms for indoor delivery tasks.

Mahjong AI

PyTorch LSTM-based neural network trained on Japanese M-League professional Mahjong rules and strategies. Integrated with web platforms via HTTP requests for real-time game analysis and move recommendations.

Day Travel Map

C++ GIS application using ezgl API for visualization. Implements A* Algorithm for optimal pathfinding, Multi-Dijkstra for multi-destination routing, and Three-Opt optimization for solving the Traveling Salesman Problem. Achieves sub-1 second response for Toronto metropolitan area routes.

Technical Skills

Languages
CC++PythonTypeScriptJavaScriptJavaC#SwiftMATLABRHTMLCSS/SCSSShell ScriptingLaTeX
Frameworks & Libraries
ReactNext.jsNode.jsROS/ROS2PyTorchTensorFlowKerasNumPyscikit-learnOpenCVOpenAILangchainThree.jsUnity
Robotics & Hardware
MuJoCoIsaac SimUR5eFrankaV-SLAMNav2ArduinoESP32PID Control
Tools & Platforms
GitDockerLinuxAWSVercelVimJupyterXCode

My Journey

A timeline of my academic and professional experience

University of Toronto

BASc Engineering Science, Robotics Specialization

Sep 2022

Full-Stack Engineer

Yuhang International Flight (YIF)

Jul 2022

UofT Math Contest

#1 All Campuses

2024

Undergraduate Research Assistant

UofT Continuum Robotics Laboratory

Feb 2024

Putnam Math Contest

Top 154 Individual, Top 5 Team

2023

Research Assistant

Huawei Technologies, Noah's Ark Lab

May 2025

Current

Full-Stack Engineer / Frontend Lead

Prelude CA

Jun 2025

Current

Expected Graduation

BASc Engineering Science

Apr 2027